Estimating Transit Ridership Patterns Through Automated Data Collection Technology A Case Study in San Luis Obispo, CA Ashley Kim ITE Western District Annual Meeting San Diego, CA June 20, 2017 1
Overview Objective Literature Review Research Design Data Filtering Results Conclusions 2
Objective 3
Objectives Evaluate the reliability of data collected with emerging automated data collection technologies BlueMAC and Automatic Passenger Counter (APC) Estimate origin-destination matrices based on the data Analyze ridership patterns in a tactical urbanism setting: Downtown San Luis Obispo Farmer s Market 6/12/17 4
Why assess origin-destination flow patterns through automated technologies? Transit route planning & stop location placement Identify travel patterns to provide conclusions for future planning, operational analysis, & service management Current data collection methods are labor-intensive and costly Ex. On-board surveys, smart card transactions, video recording Automated technologies can provide continuous data at a low cost-per-datum rate 5
Literature Review 6
Current data collection techniques Automatic passenger counter (APC) SLO Transit Automated fare collection system (AFC) Clipper Card for Bay Area, CA MetroCard for New York City, NY Onboard survey LYNX Light Rail Blue Line Extension by the Charlotte Area Transit System 7
Bluetooth data collection has been used for various transportation engineering applications Mass movements Evacuation procedures Tourism patterns Construction effects Movements through airport security Vehicular travel time Los Osos Valley Road in San Luis Obispo, CA 8
Bluetooth devices use radio waves to wirelessly connect to a phone or computer. Source: Libelium 9
Tactical Urbanism Low-cost, temporary changes to the built environment. PARK(ing) day On-street parking to public space Awareness towards allotted space for private vehicle storage Downtown San Luis Obispo Farmer s Market Thursdays, 6-9PM Street/structure parking maximized Higher transit demand 10
Research Design 11
Study Area City of San Luis Obispo 2009-2013 Population: 58,684 including Cal Poly students 99% of residents live within 0.25 mile from transit stop SLO Transit Public bus service to City and Cal Poly San Luis Obispo 7 weekday routes 6 Saturday routes 4 Sunday routes 12
Route 6A Route 4/5 Route 6B Route 1 Route 3 Route 2 Source: OTvia 13
Four data collection methods were utilized for this study. Bluetooth Data 5 BlueMAC devices February-March 2017 Detection time stamp and MAC ID GPS Probe 2 Probe Days, 3 Probe Runs March 21 and March 22, 2017 Latitude, longitude, elevation, time Automatic Passenger Counter (APC) Bishop Peak Technology Historic GPS data of buses Passenger Survey 100 passengers @ Kennedy Library Bus stop Trip frequency, wait times Bluetooth enabled? Why? 27 out of 100 passengers have Bluetooth enabled during bus trips 14
Bluetooth detection specifications Unique match media access control (MAC) ID per device Low-energy USB plug-ins for 30 ft detection radius Class I Bluetooth Source: Digiwest 15
BlueMAC Data February 28 - March 30, 2017 30 total days Five BlueMAC Devices Each deployed on a SLO Transit bus Daily Dispatch Logs to match routes BlueMAC ID Bus ID Number CP-01 858 CP-02 860 CP-05 862 CP-04 1264 DIGI-150 1365 16
Bluetooth Sensitivity Check Do passengers stay within detection range during their trip? What if an adjacent vehicle on the road is detected? What if a bicyclist, pedestrian or non-passenger is detected? 17
Distance (miles) A Bluetooth sensitivity check was conducted on April 28, 2017 on Route 4. Time-Space Diagram, Trial #1, Route 4, PM Run 2.5 2 Downtown Transit Center Mill at Santa Rosa 1.5 Mill at Pepper Mill at Grand Grand at Abbott 1 0.5 Probe BlueMAC Performing Arts Center 0 Kennedy Library 0:00 1:00 2:00 3:00 4:00 5:00 6:00 7:00 8:00 9:00 10:00 11:00 12:00 Time (mm:ss) 18
Data Filtering 19
The data was binned into projects by week and route, then downloaded as CSV files 20
The BlueMAC website displayed the Capture Rate of each device: the number of devices detected per hour. 21
Hour of Day APC and Unfiltered BlueMAC Data Raw Data from Route 2 on Tuesday, February 28, 2017 APC Counts & Bluetooth devices detected 0 50 100 150 200 250 300 350 400 450 500 17 16 15 14 13 12 11 10 9 8 7 6 APC Bluetooth 22
Compared to the APC passenger counts, there were significantly more Bluetooth devices detected due to noise and inconsistencies. Day Date Route Bluetooth APC Percent Difference (APC BT)/APC Tuesday 2/28/2017 2 4,244 282-1404.96% Wednesday 3/1/2017 2 4,169 291-1332.65% Thursday 3/2/2017 2 4,317 280-1441.79% Tuesday 3/7/2017 3 1,613 426-278.64% Wednesday 3/8/2017 3 1,580 417-278.90% Thursday 3/9/2017 3 1,712 390-338.97% Tuesday 3/14/2017 6A 5,521 970-469.18% Wednesday 3/15/2017 6A 4,836 976-395.49% Thursday 3/16/2017 6A 5,283 921-473.62% Average Percent Difference -712.69% 23
24
Number of Devices Number of Devices After filtering the data, the weekly total detected devices were plotted with the raw data. Bluetooth-Detected Devices on Route 2 With Filtered Data 16,000 14,000 12,730 12,000 10,000 8,000 6,000 4,000 2,000 13,787 Raw Data After Filtering 1,694 1,428 0 Week 1 Week 3 Bluetooth-Detected Devices on Route 5 With Filtered Data 18,000 16,000 14,000 12,000 10,000 8,000 6,000 4,000 2,000 0 14,818 14,992 15,867 609 870 509 Week 1 Week 2 Week 4 Raw Data After Filtering 25
Hour of Day Hour of Day APC and BlueMAC hourly counts for Tuesday, Wednesday, and Thursday were plotted for Routes 3 and 6A. 17 16 15 14 13 12 11 10 9 8 7 6 Filtered Data from Route 3 on Tuesday, March 7 APC Counts & Bluetooth devices detected 0 10 20 30 40 50 60 APC Bluetooth 20 19 18 17 16 15 14 13 12 11 10 9 8 7 Filtered Data from Route 6A on Tuesday, March 14 APC Counts & Bluetooth devices detected 0 20 40 60 80 100 120 APC Bluetooth 26
Hour of Day For Route 2, hourly Bluetooth device counts were higher than APC hourly counts. Filtered Data from Route 2 on Tuesday, February 28, 2017 APC Counts & Bluetooth devices detected 0 10 20 30 40 50 60 70 80 17 16 15 14 13 12 11 10 9 8 7 6 APC Bluetooth 27
Devices Detected The data filter for trip duration was adjusted on statistical analysis software to different trip durations for Route 2. 40 Bluetooth Device Counts With Different Filters from Route 2 on Tuesday, February 28, 2017 35 30 25 20 15 15 minutes 20 minutes 25 minutes 30 minutes APC 10 5 0 6 7 8 9 10 11 12 13 14 15 16 17 Hour of Day 28
Overall, Route 6A had the most consistent percent differences, and was chosen for the OD estimation. Day Date Route Bluetooth Raw Detection Bluetooth Filtered Detection APC % Difference (APC-BT)/APC Tuesday 2/28/2017 2 4,244 512 282-81.56% Wednesday 3/1/2017 2 4,169 528 291-81.44% Thursday 3/2/2017 2 4,317 654 280-133.57% Tuesday 3/7/2017 3 1,481 200 407 50.86% Wednesday 3/8/2017 3 1,580 419 417-0.48% Thursday 3/9/2017 3 1,551 270 386 30.05% Tuesday 3/14/2017 6A 5,521 678 970 30.10% Wednesday 3/15/2017 6A 5,577 766 1,083 29.27% Thursday 3/16/2017 6A 5,283 678 921 26.38% Average Percent Difference -14.49% 29
Results 30
Origin-destination matrices were generated for Route 6A. Cal Poly 31
APC data for Route 6A was used to generate an OD matrix. 3/16/17 5:10pm and 5:40pm runs from APC data RouteID Route Stop ID StopName Observed CountIn CountOut 957 Route 6A 63 Highland at Mt. Bishop 3/16/2017 9:06 0 1 957 Route 6A 63 Highland at Mt. Bishop 3/16/2017 9:06 0 1 957 Route 6A 63 Highland at Mt. Bishop 3/16/2017 9:06 1 0 957 Route 6A 63 Highland at Mt. Bishop 3/16/2017 9:06 1 0 Number of Rides by Destination Cal Poly Kennedy Library Highland at Mt. Bishop Patricia Highland Highland Patricia at at at Cuesta at Jeffrey Highland Foothill Foothill at Chorro ROUTE 6A Foothill at Ramona at S. Ramona at Casa at Origin La Entrada Tassajara Palomar Murray Cal Poly Kennedy Library - S - 0 21 2 3 2 0 13 9 2 0 0 Highland at Mt. Bishop 0-0 0 0 0 0 0 0 0 0 0 Highland at Cuesta 8 0-0 0 0 0 0 0 0 0 0 Highland at Jeffrey 1 0 0-0 0 0 0 0 0 0 0 Patricia at Highland 3 0 0 0-0 0 0 0 0 0 0 Patricia at Foothill 1 0 0 0 0-0 0 0 0 0 0 Foothill at La Entrada 0 0 0 0 0 0-0 0 0 0 0 Ramona at S. Tassajara 1 0 0 0 0 0 0-0 0 0 0 Ramona at Palomar 7 0 0 0 0 0 0-0 0 0 Foothill at Chorro 0 0 0 0 0 0 0 0 0-0 0 Casa at Murray 1 0 0 0 0 0 0 0 0 0-0 Casa at Deseret (NB) 0 0 0 0 0 0 0 0 0 0 0 - Casa at Deseret (NB) 32
APC data was used to compare ridership for Route 6B on Tuesday versus Thursday. Cal Poly 33
APC Count In APC data for Route 6B connecting Cal Poly to Downtown SLO was analyzed for Thursday PM periods for Downtown San Luis Obispo Farmer s Market. 350 APC Data for Multiple Tuesdays and Thursdays on Route 6B 300 250 200 150 100 50 0 5:00 PM 6:00 PM 7:00 PM 8:00 PM 9:00 PM 10:00 PM Time Farmer's Market Tuesday Thursday 34
An OD matrix was generated using APC data for Route 6B comparing Tuesday and Thursday PM counts. Tues. & Thurs, 3/14 & 3/16 5:00pm to 11:00pm from APC Number of Rides by Destination Mill at Phillips at Mill at Mill at Mill at Mill at Grand at Grand at Mill at Pepper Pepper Johnson Mill at Santa Downtown Santa Rosa Johnson Phillips at Pepper California California Origin Cal Poly McCollum Wilson Park (WB) (WB) (WB) Rosa (WB) Transit Center (EB) (EB) Pepper (EB) (EB) at Phillips at Taft Cal Poly - 33, 26 40, 67 11, 8 11, 9 7, 3 12, 5 24, 8 28, 94 0 0 0 0 0 0 Grand at McCollum 0-0 0 0 0 0 0 1 0 0 0 0 0 0 Grand at Wilson 0 - - 0 0 0 0 0 4 0 0 0 0 0 0 Mill at Park 0 - - - 0 0 0 0 0 0 0 0 0 0 0 Mill at Pepper (WB) 0 - - - - 0 0 0 1 0 0 0 0 0 0 Phillips at Pepper (WB) 0 - - - - - 0 0 2 0 0 0 0 0 0 Mill at Johnson (WB) 0 - - - - - - 0 0 0 0 0 0 0 0 Mill at Santa Rosa (WB) 0 - - - - - - - 6 0 0 0 0 0 0 Downtown Transit Center 32, 103 - - - - - - - - 25 6 1 1 8 1 Mill at Santa Rosa (EB) 6, 7 0 0 0 0 0 0 0 0-0 0 0 0 0 Mill at Johnson (EB) 5, 1 0 0 0 0 0 0 0 0 0-0 0 0 0 Phillips at Pepper (EB) 1, 2 0 0 0 0 0 0 0 0 0 0-0 0 0 Mill at Pepper (EB) 1, 1 0 0 0 0 0 0 0 0 0 0 0-0 0 California at Phillips 2, 0 0 0 0 0 0 0 0 0 0 0 0 0-0 California at Taft 1,1 0 0 0 0 0 0 0 0 0 0 0 0 0-35
Conclusions 36
BLUETOOTH Noise and inconsistencies Sample size: ~12% Probe run check Data filtering required Trip durations from detection times Passenger effort: enable Bluetooth Privacy implications from MAC IDs $3,200 per device No effort from driver No onboard survey Automated, wireless data collection Low costs Short installation time Minimal maintenance APC Exact passenger counts at each bus stop Sample size: ~100% Specific passenger routes must be inferred Passenger effort: board and exit the bus Passenger identities are anonymous $2,000 per bus door 37
Downtown Farmer s Market: the APC data showed higher ridership on Thursdays versus Tuesdays 38
Origin-destination estimation Time consuming for both APC and BlueMAC Similar trends for APC and BlueMAC Future studies Longer data collection period for specified routes Use statistical analysis software to count ODs Group key locations into zones 39
Further Analysis and Research Data collection methodology Optimal hardware settings and placement Second detector on bus Detector at bus stop Data collection time period Explore other methods: Wi-fi Compare other citys data Data filtering procedures Specific data filter for specific route Test multiple iterations of filters Explore different statistical analysis software OD Estimation Pair bus GPS coordinates to BlueMAC time stamps onto GIS Map 40
Acknowledgements Thesis Committee Dr. Anurag Pande, Advisor Dr. Kimberley Mastako, Committee Dr. Cornelius Nuworsoo, Committee Cal Poly Civil & Environmental Engineering Department City of San Luis Obispo Public Works Transportation Engineering Team Bryan Wheeler SLO Transit & Bishop Peak Technology Dee Lawson Gamaliel Anguiano John Osumi Cal Poly Statistics Department Mitchell Collins Professor Rebecca Ottesen Digiwest Cal Poly ITE friends & 2016-2017 Officer Team Transportation Engineering Student Project Area Friends & Family 6/12/17 41
Thank you for listening! Ashley Kim, EIT akim56@calpoly.edu